[MakerFlyFly]

Summary

• Keep contributor PRs in the untriaged backlog when their only engagement is their author's own label or draft-state event.
• Limit the filter to LabeledEvent, ReadyForReviewEvent, and ConvertToDraftEvent; preserve collaborator activity, closure actors, ready-label behavior, and label timestamps.
• Make the reduced closed-PR fixture retain an explicitly supplied timeline and verify the dashboard metrics and written JSON sidecars. Include the identical generated dependency copy required by the plugin packaging hook.

Type of change

• Python package (tools/pr-management/)

Test plan

• On original upstream b6ccd434, six contributor regression cases fail and the other 30 targeted cases pass.
• With the fix, all 116 stats tests pass.
• Full affected package: 795 tests and 15 subtests pass on Linux with Python 3.12.12.
• Full-package Ruff, format check, and Mypy pass.
• The four changed file blobs match the contributor-reviewed final patch.
• prek run --all-files --skip workspace-pytest passes, including lychee and the generated-plugin check.
• Scoped workspace pytest passes for both affected members: tools/pr-management has 795 passed and 15 subtests; plugins has 114 passed and one container-permission probe skipped. Commit hooks were not bypassed.
• Magpie pre-first-PR: ready, zero blocking and advisory findings.

RFC-AI-0004 compliance

This corrects deterministic classification without adding fetched fields, configuration, dependencies, or service access.

Linked issues

Closes #2015.

Notes for reviewers

The PR is one commit based on upstream main@b6ccd434.
The four reviewed source/test blobs are unchanged; a fifth file is the generated magpie-security-tools copy of the classifier, byte-for-byte identical to the reviewed source.
A currently draft PR remains outside the untriaged non-draft counts.

The label case is defensive input coverage; ordinary external authors need repository triage permission to apply labels.
Closed-PR assertions deliberately retain legacy behavior within this repair's scope; they do not endorse contributor self-closure as maintainer engagement.
Strict actor authorization and comment/review engagement policy remain unchanged.

Suggested labels: bug, family:pr-management, family:tools, capability:stats, capability:fix, substrate:analytics.

Was generative AI tooling used to co-author this PR?

• Yes — OpenAI Codex (GPT-6) assisted with implementation, tests, and drafting.
• The contributor reviewed the final implementation before upstream submission.
• The commit carries Generated-by: Codex (GPT-6).

[Kaap10]

LGTM! Thank you for the clean fix and thorough test coverage.

• Accurately addresses #2015 by ensuring contributor self-events (ReadyForReviewEvent, ConvertToDraftEvent, LabeledEvent) do not falsely trigger maintainer engagement.
• Safely handles casing discrepancies and deleted user accounts with (author or "").lower().
• Preserves maintainer actions on contributor PRs and collaborator self-events.
• Full test suite passes cleanly with parity across both tool locations.